Quoted By:
Okay so after my latest lora test, 103 videos from a single anime, I feel like the style is closer, but didn't quite reach the point I needed, increasing the strength seems to help so leads me to think I need more data on characters/steps. Thinking about it more, I kinda want to try a character lora... I was thinking about scraping any 3d streams bae has and using those as my dataset, sounds like a night of captioning and splitting videos again (I am of a rational mind and not slipping into madness)
el psy kongroo
